This application claims the equity for the U.S. Patent application 15/061,202 that on March 4th, 2016 submits, should by referenceU.S. Patent application is incorporated herein with its entirety.
Specific embodiment
In order to limit the purpose of present disclosure, " Glucose results measured " are the people such as measured by glucose sensorGlucose level；" actual glucose is horizontal " is that the actual glucose of people is horizontal；And " glucose level of estimation " is peopleEstimation glucose level, be potentially based on the Glucose results measured.
With reference to Fig. 1, diagram is for monitoring the exemplary continuous grape for the glucose level for suffering from person with diabetes (PWD) 11Glucose monitor (CGM) system 10.Particularly, CGM system 10 can be operated for predetermined adjustable interval (such as each minute, every fiveMinute) or with other appropriate intervals collect the dextrose equivalent that measures.The CGM system 10 is illustratively included with insertion people's 11The glucose sensor 16 of spicule or probe 18 below skin 12.The end of the spicule or probe 18 is located at interstitial liquid streamBody 14(such as blood) or another body fluid in so that being based on interstitial liquid stream by the measurement result that glucose sensor 16 obtainsThe level of glucose in body 14.The glucose sensor 16 is positioned adjacent to the abdomen of people or in another appropriate location.In addition, shouldGlucose sensor 16 can be calibrated regularly to improve its accuracy.The periodic calibration, which can help to correct, is attributed to sensingVariation in the physiological status of sensor drift and sensor insertion site that device degrades.Glucose sensor 16 can also wrapInclude other component comprising but it is not limited to wireless transmitter 20 and antenna 22.Glucose sensor 16 can be alternatively using useIn other the appropriate equipment for obtaining measurement result, such as noninvasive equipment (such as infrared light transducer[sensor).Obtaining measurement knotWhen fruit, glucose sensor 16 transmits the dextrose equivalent measured, the calculating equipment 26 to equipment 26 is calculated via communication link 24Illustratively blood glucose (bG) management equipment 26.The bG management equipment 26 can be configured to will be whithin a period of time from grapeMultiple Glucose results measured that sugared sensor 16 receives and the counterpart impedance measurement from impedance measuring sensor 25As a result it is stored in memory 39.
The CGM system 10 further comprises the therapy delivery device 31 for therapy (such as insulin) to be delivered to people(illustratively insulin infusion pumps 31).Infusion pump 31 is communicated via communication link 35 with management equipment 26, and management equipmentPill (bolus) and basal rate information can be communicated to infusion pump 31 by 26.Infusion pump 31 includes conduit 33, has and is used forThe spicule that the skin 12 by PWD 11 of insulin injection is inserted into.Infusion pump 31 is illustratively adjacent to the abdomen of people or anotherThe positioning of one appropriate location.Similar to glucose sensor 16, infusion pump 31 further includes wireless for communicating with management equipment 26Transmitter and antenna.Infusion pump 31 can operate (such as continuous or repeat release with basal rate for delivering basal insulinLow dose of insulin) and bolus insulin (such as insulin of surge dosage, such as near canteen event).It can ringYing Yu inputs or responsive to the order from management equipment 26 by the user that user triggers to deliver bolus insulin.It is similarGround inputs based on user or is arranged in response to the order from management equipment 26 basal rate of basal insulin.Infusion pump31 may include for showing the display 37 of Pump data and providing the user interface of user's control.In an alternative embodimentIn, infusion pump 31 and glucose sensor 16 may be provided as the individual equipment worn by patient, and by processor or micro-At least part for the logic that controller provides may reside on the individual equipment.It can also be (such as logical by other devicesIt is manual via spicule to cross user) injection bolus insulin.
Communication link 24,35 is illustratively wireless, such as radio frequency (" RF ") or other wireless frequencies appropriate, at itIn via electromagnetic wave data and control are transmitted between sensor 16, therapy delivery device 31 and management equipment 26." bluetooth "It is a kind of wireless RF communication system of exemplary types, uses the frequency of about 2.4 gigahertzs (GHz).Another exemplary classThe radio communication system of type uses infrared light, the system such as supported by Infrared Data Association (IrDA).It can be providedThe wireless communication of his appropriate type.In addition, each communication link 24,35 can promote (such as glucose between multiple equipmentBetween sensor 16, calculating equipment 26, infusion pump 31 and other equipment or system appropriate) communication.It is alternatively possible to beingWire link, such as wired ethernet link are provided between the equipment of system 10.It is appropriate public or special that other can be usedWith wired or wireless link.
The exemplary bG management equipment 26 of the CGM system 10 of Fig. 2 pictorial image 2.Management equipment 26 includes executing to be stored in pipeManage at least one microprocessor or microcontroller 32 of the software and/or firmware code in the memory 39 of equipment 26.The software/Firmware code includes to promote management equipment 26 to carry out function described herein when the microcontroller 32 of managed devices 26 executesThe instruction of energy.Management equipment 26 can alternatively include one or more specific integrated circuits (ASIC), field-programmable gate arrayArrange (FPGA), digital signal processor (DSP), firmware hardwired logic or combinations thereof.Although management equipment 26 is illustratively grapeGlucose monitor device 26, but other appropriate managerial equipment 26 can be provided, such as desktop PC, calculating on kneeMachine, computer server, personal digital assistant (" PDA "), smart phone, cellular device, tablet computer, infusion pump includingThe integrated equipment of glucose measurement engine and PDA or mobile phone etc..Although management equipment 26 is illustrated as single management equipment 26,But multiple calculating equipment can be used together to execute the function of management equipment 26 described herein.
Memory 39 is any computer-readable medium appropriate that can be accessed by microcontroller 32.Memory 39 can beSingle storage equipment or multiple storage equipment can be located inside or outside management equipment 26, and may include volatibility andBoth non-volatile medias.In addition, memory 39 may include that can be removed and one or two in non-removable media.ExampleProperty memory 39 include random-access memory (ram), read-only memory (ROM), electrically erasable ROM(EEPROM),Flash memory, CD-ROM, digital versatile disc (DVD) or other optical disk storage apparatus, magnetic storage device are configured toStoring data and can managed devices 26 access any other medium appropriate.
Management equipment 26 further comprises the operably coupled communication equipment 50 to microcontroller 32.Communication equipment 50 includesCan operate any appropriate wireless and/or wire communication module for transmitting and receiving data and by communication link 24,35 are controlled between equipment 26 and glucose sensor 16 and infusion pump 31.In one embodiment, communication equipment 50 wrapsInclude antenna 30(Fig. 1 for wirelessling receive and/or transmitting data by communication link 24,35).Management equipment 26 will measureGlucose results and stored via communication equipment 50 from glucose sensor 16 and/or received other data of infusion pump 31In memory 39.
Management equipment 26 includes (one or more) user input equipment 34 for receiving user's input.(one or moreIt is a) input equipment 34 may include button, switch, mouse pointer, keyboard, touch screen or any other input equipment appropriate.Display 28 is operatively coupled to microcontroller 32, and may include being configured to show the information provided by microcontroller 32To any display appropriate or monitor technology (such as liquid crystal display etc.) of user.Microcontroller 32 be configured toDisplay 28 transmission with the related information of glucose state of detected people, risk associated with glucose state andBasal rate and pill information.The glucose state may include the gentle estimated glucose level of estimated glucose waterChange rate and estimated glucose level quality or probabilistic estimation.In addition, shown information can wrapInclude about the glucose level of people that is estimated or being predicted be hypoglycemia or hyperglycemia warning, notice etc..For example,If the glucose level of people falls on (such as every deciliter 50 to 70 milligrams of blood of (or being predicted to fall on) predetermined hypoglycemia threshold value(mg/dL) glucose) hereinafter, can then give a warning.Management equipment 26, which can be configured to convey in tactile to people, to be believedBreath or warning, such as pass through vibration.
In one embodiment, management equipment 26 with such as at the facility of care-giver or the addressable position of care-giverRemote computing device (not shown) communication, and between them transmit data (such as glucose data or other physiology letterBreath).In this embodiment, (such as via internet, management equipment 26 and remote equipment are configured to through data connectionCellular communication) or such as disk, usb key, compact disk or other portable memory devices etc memory devices objectReason transmitting is to transmit physiologic information.
Microcontroller 32 includes gain control module 43, is based on via open loop glucose control 45 or closed loop glucoseThe insulin delivery of 47 control of control sends the programmed logic that the glucose state of people is maintained to target glucose state.RecognizeIt arrives, which grasps together via communication link 35 and open loop glucose control 45 or closed loop glucose control 47Make, because it is by gain control module 43 via receiving signal from glucose sensor 16 and/or impedance transducer 25The total mass fraction Q calculated when inputtotalAnd it selects and regularly updates.It is target glucose state description not have phaseOptional or ideal glucose state (the gentle glucose zero of the glucose water of such as 112.5 mg/dl of associated harm or riskChange rate), although any target glucose state appropriate can be identified.What is provided in subsequent chapters and sections below illustrative showsIn example, gross mass score is to promote microcontroller 32 to receive about from glucose sensor 16 based on gain control module 43(one or more) signal input characteristics and data (i.e. about multiple Glucose results measured) and come from impedance transducer25 associated impedance value and the analysis run, and open loop glucose control is selected using the information obtained from analysis resultSystem 45 or closed loop glucose control 47 and the such information of offer is current with directive therapy delivery device 31 on the display 28PWD 11 is just operated in which kind of mode (open loop or closed loop).It discusses by coming from glucose sensor 16 and/or impedance transducerThe information that 25 signal input provides.
Fig. 3 A describes from the figure for being coupled to the Glucose results 40 that the glucose sensor 16 for suffering from person with diabetes measuresThe example of table.Shown great circle 42 shows the dextrose equivalent measured using standard blood sugar measuring system, such as can by withIn the calibration of continuous glucose sensor.The connecting line of small circle 44 is shown by subcutaneous glucose sensor (such as sensor 16(Fig. 1)) dextrose equivalent that measures.Until by 1.25(place sensor until number of days) instruction time after soon whenBetween until, the track reference value well of sensor 16 illustrates good performance.During this time period, it is obtained using sensor 16To impedance measurements the pattern seen in figure 3b is shown.Solid line 46 and 48 is shown respectively discribed to be measured by sensorImpedance magnitude " Zmod " (by ohmmeter) and phase angle " Zphz " (by spend in terms of) and frequency (Hz) between relationship.With regard to magnitude andThe result for making the equivalent-circuit model of impedance data fitting description sensor 16 is shown respectively in the dotted line 47 and 49 of phase angle superposition.It shouldModel is fitted from 1Hz to 10 kHz with the data measured good.The fitting result is the instruction of 16 operational excellence of sensor, and withThe ratio of reference value in Fig. 3 A is more consistent.It is (wherein good unanimously to intend (lack-of-fit) value with small mistake by modeling statisticsIt is associated) in mistake intend parameter (" error ") or the error as shown in the illustrated examples of following table 1 to quantify the number measuredAccording to the consistent degree between model.
Table 1:
。
In the illustrative embodiments of Fig. 3 A, after 1.25 days time soon, sensor 16 breaks down and showsUnreasonable value.Fig. 3 C is shown the magnitude just measured again in the impedance measurements at this moment obtained using sensor and measuredPhase angle described respectively with line 46,48.With line 46,48 describe the value measured and for being described with line 47,49 it is equivalentThe fitting of circuit model has significant deviation with regard to model magnitude and model phase angle respectively.The deviation is unreliable sensor performanceIndicator, and parameter (" error ") is intended to measure by the mistake in modeling statistics shown in the illustrated examples such as following table 2Change.
Table 2
,
In view of foregoing, wherein impedance and/or phase angle measurement result no longer with the known equivalents circuit model of sensorThe consistent sensor of form can be determined to be breaking down and be indicated as not delivering authentic data.In some implementationsIn example, such figure description of the information illustrated by Fig. 3 A-3C can be supplied to display by microcontroller 32 and (such as mentionedSupply one or two of display 28 and 37) for consulting (if necessary) by user.
It would be recognized that the biggest obstacle for insulin delivery system using Automatic Control Strategy has been from (oneIt is a or multiple) uncertainty of the input data (i.e. dextrose equivalent) of continuous glucose sensor.Improve accuracy andVery big effort is made in terms of reliability.For example, even if being adjusted using used impedance data from glucose sensorData are also less likely to realize 100% reliability to improve the accuracy of report value.Because of the reason, gain control module 43The quality of sensing data is assessed, and considers the quality in control algolithm.Specifically, inventor has found from sensorThe quality of report value can be improved in the probability score that the time course of data exported and be used to adapt to the gain of statistical zero-knowledge.
Because of reason above and according to each embodiment disclosed herein, embodiment described herein is generallyIt is related to the method and system delivered for insulin, and relates more particularly to a kind of controller for insulin delivering, itRealize controller gain based on probability, based on the quality calculated of the signal inputted from glucose sensor so as to more preferableGround determines the insulin delivering for maintaining the good glucose control of people required.Specifically, estimation is described according to from one groupThe gross mass score of quality metricThe data and passed from glucose that (such as (one or more) signal input characteristics) is assessedThe system and method for the quality for the data that sensor 16 and/or impedance transducer 25 receive.Gross mass scoreWith from zero(0) value of (one) 1 range is arrived, and is used to change the gain of microcontroller 32 to provide increasing when quality score is lowBig level of security, and microcontroller performance and glycemic control are improved when quality score is high.
For example, having in an illustrative realization in sensor 16=0(or 0≤ ≥T)In the case where gross mass score, if microcontroller 32 is based on institute by microcontroller 32 wherein with closed loop glucose control 47(The input from sensor 16 and 25 that receives is automatically performed insulin delivering adjustment) it operates, then microcontroller 32 willReturn to safer operated open-loop mode (i.e. open loop glucose control 45(Fig. 2)), in open loop glucose control 45 by withIt family and completes to adjust as the time section by pre-programmed is recommended to the user.In gross mass scoreGreater than thresholdValueTAnd/or it is equal to one (1) (i.e.T <≤ 1) in the case where, microcontroller 32 more energetically takes action determining biographySensor data will manage blood glucose in the case where the insulin for enabling a system to delivering appropriate amount, to allow such as system 10 with mostGood safe mode is run in best closed loop glucose control 47.In some embodiments, in memory by threshold valueTIt is presetIt is accurate enough at indication sensor data so that system can deliver the insulin of appropriate amount high probability (such asT=0.9)Value, and in other embodiments can be by PWD 11 or health care worker by threshold valueTIt is arranged and/or is adjusted to enoughThe particular value of threshold value is provided between the open loop and closed-loop control for PWD 11.In a specific embodiment, work as microcontrollerWhen therapy delivery device 31 is returned to closed-loop control from opened loop control automatic switchover by device, whenWhen close to 1 or at 1, by micro-Controller 32 for example notifies user on display 37.
Fig. 4, which describes, is used to determine that the best insulin for the good glucose control that will maintain PWD 11 delivers control (i.e.Open loop or closed-loop control) gain control module 43 each model and input to gain control module 43.The gain controlMolding block 43 receives the Glucose results 40 measured from the glucose sensor 16 for being coupled to PWD 11 and carrys out self-impedanceThe data of sensor 25.It is gentle that the glucose sensor 16 and impedance transducer 25 may be configured to periodic measurement glucose waterThe impedance of people and the Glucose results 40 measured and associated impedance measurements 41 are sent to gain controller model43(is for example via communication link 24).It can be with by period of the sensing and transmission of glucose sensor 16 and impedance transducer 25It is the period (such as 1,5 or 10 minute) of selection in from 1 to 10 minute, and can be by PWD 11 or health care worker's exampleAs the user via Blood sugar management equipment 26 inputs the setting of 34 factories and/or adjustable setting.
As described by Fig. 4, which may include probability analysis tool 54 and recursion filter 52.It shouldProbability analysis tool 54 may be configured to receive the Glucose results 40 measured and sensed by impedance transducer 25 andImpedance data 41(such as AC impedance associated with each Glucose results 40 measured).The probability analysis tool54 are then based on the Glucose results 40 measured received in this way and associated impedance data 41 carrys out matter from one group of calculatingMetric determines gross mass score61.For example, such quality metric can be from the Portugal measured receivedThe probability for the glucose sensor accuracy that grape sugar result 40 calculates58.The probability of glucose sensor accuracy 58It is that glucose sensor 16 operates normally (i.e. there is no failures) and the probability value of calculating.Jointly owned United States Patent (USP) 8,843,321 discuss the probability that glucose sensor accuracy how can be calculated from the Glucose results measured receivedThe disclosure of which is completely integrated herein by 58(by quoting).Determining gross mass scoreDuring useIt is in quality metric the other is derived from AC impedance measurements sensor mass score59, i.e. glucose sensor16 suitably contact PWD 11 and provide the probability of non-erroneous sensor values.Gross mass score61 can also use withOther related measurements of the quality of sensor measurement.Therefore, output (the i.e. gross mass score of probability analysis tool 5461) being used to distinguish from sensor noise and poor sensed characteristic, (each of which may have normal state pointCloth) and sensor failure (it may not be normal distribution), and by with for choosing as discussed previously aboveThe operation therapy delivery device 31 in which kind of control model is selected, i.e., or open loop glucose controls 45 or the control of closed loop glucose47.Hereafter it is discussed below how probability analysis tool 54 exports gross mass score61。
The probability analysis tool 54 may include that any number of can analyze the Glucose results 40 measured, AC impedanceMeasurement result 41 and/or its variation simultaneously calculate minimum gross mass score from quality metric group above-mentioned61Mathematical algorithm.For example, using hidden Markov model come based on the Glucose results 40 and AC impedance measurements 41 measured comeEstimate minimum gross mass score61.As discussed above, the output of probability analysis tool 54 has pair of the range from 0 to 1InValue (probability that itself and sensor operate normally is related and is microcontroller 32 in closed loop glucose control 47Operation provides the data of safe enough), and the insulin of appropriate amount is delivered to PWD 11 to tie up the glucose state of peopleIt holds in target glucose state.The probability analysis tool 54 can be configured to receive gross mass score61 can be with baseIn other kinds of data, when such as people has eaten a meal, when people takes exercise and when to people deliver insulin.ThisIt is discussed below now afterwards and how to export sensor mass score59。
The sensor mass score59 can export from AC impedance measurements (AC), the AC impedance measurementsIt (AC) is that offer and losing quasi- of the value measured using the assessment of known equivalent-circuit model is passed through by impedance measuring equipment 25.ExampleSuch as, " error " (E) can be used in following equation:
= 0.5*(1-tanh(a*log(b*E)),
To calculate in 1(for low-down E) and zero (for high E) between value.It can choose parameteraWithbTo choose 0.5Value (1/b) position and 0.5 value (a) at change rate.Alternatively,Value can be from making the value be fitted to equivalent electricityRoad model and the export of parameter that result obtains.Parameter provides the sensor compared with the batch value derived from its previous or factoryThe quantitative comparison of electrochemical properties.It can will sum it up with the deviation of predetermined value into error term, and be formed such as feelings in frontThe error term calculated in condition.Then make the sensor mass score derived from AC impedance measurements59 with58 valuesCombination, and gross mass score is calculated by finding out minimum mass score from quality metric group according to following equation61:
。
It would be recognized that output (the i.e. gross mass score of probability analysis tool 54It 61) can be using many not similar shapesFormula, such as state machine, Bayesian model or other algorithms.In one embodiment, which can be using letterThe form of single state machine, wherein gross mass score61 and glucose sensor accuracy probabilityIt can be totalBe in set { 0,1 } (i.e. depend on state machine state,61 HesEach of 58 be all 0% or 100%).?In the example and as above with reference to patent disclosed in, for the probability of glucose sensor accuracyFor, ifΔ CG(is the variation in the Glucose results currently measured according to the Glucose results previously measured) it is less than a certain negative threshold value, which will transfer to the state of sensor inaccuracy, and if Δ CG is greater than a certain positive threshold value, orIf sensor CG value (Glucose results currently measured) physiologically possible dextrose equivalent (With) in andIt has been measured in those years since the transition of the state to sensor inaccuracy rises, then sensor is branched back toAccurate state.This can mathematically be represented as:
If, then
IfOr (And), then,
If none in these jump conditions meets, state machine may remain in its current state.This is only probability pointOne example of the form of 54 adoption status machine of analysis tool.The probability analysis tool 54 can also use other forms.
In another embodiment, which may include with two states for glucose sensorHidden Markov model: 1) glucose sensor is accurate state wherein, with "SA" indicate and 2) Portugal whereinGrape sugar sensor be inaccuracy state with "SI" indicate.The hidden Markov model can provide restriction from stateSATransitionTo stateSIProbability status transition function, such as below function:
Herein "CG" it is the Glucose results currently measured, "ΔCG" it is from the Glucose results previously measured to currently measuringGlucose results variation, and α1To α4It is the constant of the characteristic depending on glucose sensor.Output for the functionThe range of value is 0 to 1,0 indicates 0% probability of sensor accuracy and 1 indicates the 100% general of sensor accuracy hereinRate.The minimum value and number 1(that " minimum " function uses mathematic(al) representation are i.e. 100%).The transition function can be based on current CGWith Δ CG value.In addition, the transition function can be (sigmoid) of S-shaped, wherein parameter alpha1And α3The position of S-shaped transition is controlled,And parameter alpha2And α4Control the slope of S-shaped.Specific people and/or sensor batch these parameters can be tuned.
The example for continuing hidden Markov model, is still in stateSIProbability (be in current stateSIWhen) may is that
And this is only Δ CG value and is in or is transitioned into stateSIPrior probabilityFunction.For the defeated of the functionThe range being worth out is 0 to 1,0 indicates 0% probability and 1 100% probability of expression herein." maximum " function uses mathematic(al) representationMaximum value and number 0(i.e. 0%).Parameter " γ " is if being less than 1 and being designed to not from CG and Δ CG value to still existingSIIn evidence then the state of hidden Markov model is gradually transitted back intoSAAttenuation term.Parameter γ can be constant and work asΔ CG it is relatively normal when can with still existSIIn probability correlation.For example, γ might be chosen such that when Δ CG is relatively normalWhen hidden Markov model still existSIIn reach about 10 minutes.The probability function further include detection CG signal in return toSAPhaseThe associated sigmoid function risen rapidly.Parameter alpha5Control the position of S-shaped transition, and parameter alpha6Control the slope of sigmoid function.It is rightIt can be tuned in specific people and/or sensor batch the two parameters.
It is depending on current stateSAOrSI, it is transitioned intoCurrent probabilityIt isOr.Glucose sensor(it is for inaccuracy) current probabilityIt can be.It is to be noted, that when being in the stateState when middle (Or) it is " 1 ", and be otherwise " 0 ".This includes being transitioned intoProbability () provide and be in'sProbability, and still existIn probability multiplied by being currently located inProbability.Value be equal to, and sensorIt is that accurate probability is simply.Therefore, for the example, the probability of glucose sensor accuracy be can be。
Fig. 5 describes two transition functionsWithGraphical representation (i.e. 1-on the histogram of Δ CG,It is in current stateShi CongIt is transitioned intoProbability).The histogram includes having centered on zero and is transitioned into sensingDevice failure and the gaussian-shape component 68 for leaving associated two tail portions of sensor fault.Two transition functions are plotted in directlyIt can be tuned with showing them to be triggered on the tail portion of histogram on side's figure.Gaussian-shape component 68 can indicate can be in grapeThe range of the Δ CG value occurred during the normal operating of sugared sensor.For example, the Δ CG value being located inside gaussian-shape component 68 canIt is attributed to sensor noise.Positioned at the outside of gaussian-shape component 68 and the Δ CG value that is located on the left of it can be attributed to sensor fromSAIt is transitioned intoSI.The shape of the distribution can be used to characterize a batch glucose sensor after manufacturing and be used to pairSensor coding.That is, transition function (With) can be adjusted to (by adjusting α1To α6Parameter) with it is rightIt is corresponding in the gaussian-shape component 68 of the glucose sensor of a particular batch.Therefore, hidden Markov model can be used toThe Glucose results measured and its variation are based only on to determine that sensor is accurate probability。
Fig. 6 illustrate show for example be attributed to glucose sensor 16 with the contact of the difference of PWD 11 in the period 70The presence of glucose sensor failure, the glucose sensor noise in the period 71 and the reduced sense in the period 72Exemplary many figures of the operation of hidden Markov model during mass metering.It in some embodiments, can be by microcontroller32 by such figure description of the information illustrated by Fig. 6 separately or together (as depicted) be supplied to display (such asIt is supplied to one or two in display 28 and 37) for consulting (if necessary) by user.
In Fig. 6, uppermost " glucose " figure includes the glucose level (dotted line) 60 for the estimation for being covered with peopleThe Glucose results (solid line) 40 measured are measured the trace of Glucose results 40 by sensor 16 from the people.In addition in Fig. 6,The gross mass score that drawing result below obtains61 and make its " glucose " with Glucose results 40 in timeTrace alignment.During discribed period time 70, glucose sensor 16 may have failure, i.e., such as by hidden MarkovWhat model determined58 may during period 70 from about 100%(such as by period 70 before and after and then exist" 1 " in " quality " y-axis indicates) it is reduced to 0%(or zero value as indicated on the quality axis) near, therefore promote simultaneouslyThe Glucose results 40 measured become inaccuracy.The instruction is attributed to the Glucose results 40 that start to measure in the period 70Value declines rapidly the detection in the hidden Markov model of (i.e. when failure occurs for the first time), is by microcontroller 32 via increasingWhat the assessment result of low or zero sensor impedance that beneficial control module 43 is made and glucose measurements quality generated.Assessment knotFruit is by " the Q that describes respectively in Fig. 6ac" and " Qg" to draw 63 and 65 indicate, and 63 describes based on measuring wherein drawingGlucose results 4058 value, and drawing 65 is based on corresponding A C impedance and/or phase angle measurement result 41QAC 59Value.
Still referring to FIG. 6, glucose sensor 16 may start the normal operating (Portugal measured at the end of period 70Grape sugar result 40 becomes accurate again), and gross mass score61 can increase again and return to about 100%.As before,This may be attributed at the end of period 70 via the sensor impedance by microcontroller 32 via gain control module 43With glucose measurements quality in 100%(or 1) nearby 100%(or 1) at the value of Glucose results 40 that measures of assessmentThe detection in hidden Markov model to increase sharply (i.e. when glucose sensor 16 is back to when normal operating).In the periodThis increases sharply by the sensor accuracy quality " Q described in Fig. 6 after 70ac" draw 63 and sensor measurement quality“Qg" drawing 65 indicates.Gross mass scoreHow promptly 61 rate of change is likely to be dependent on glucose sensorFrom breaking down, (inaccuracy) is transitioned into normal (accurate) operation, i.e. rate58 nearby change to about 100% from 0%(or 0)(or 1) andQACAlso about 100%(or 1) is kept.For example, if transition is relatively rapid,58 can be fast near 100%Speed is transitioned into about 0%.However, if glucose sensor is slowly transitioned into normal operating from breaking down,58About 100% can be slowly transitioned near 0%.If still existed from CG and Δ CG value there are seldom evidence or without evidenceSIIn, then (Being found in equation) attenuation term γ can permit58 gradually transit back intoSA。
Still referring to FIG. 6, the glucose sensor noise for being shown as occurring during the period 71 can also promote total matterIt measures point61 reduce, this depends on the seriousness and level of noise.As depicted in figure 6, in 71 period of period PortugalGrape sugar sensor noise can promote gross mass score61 is even between good (100% or 1) and poor (0% or 0) sensing qualityYou reduce.Equally, in the period 72, due to inappropriate sensor placement/contact difference sensing quality is (such as in the timeBy Q in sectionacDrawing 63 indicates usual at 0% or 0) gross mass score can also be promoted61 reduce, this depends on making an uproarThe severity and level of sound.As described in Fig. 6, total matter can be promoted in 71 period of period glucose sensor noiseIt measures point61 reduce once in a while between good (100% or 1) and poor (0% or 0) sensing quality.Certainly, glucose sensor eventBoth barrier and sensor noise and difference sensing quality can have the change level of amplitude and/or duration, such as in grapeIt is apparent in the correspondence period 70,71 and 72 of sugared result 40.In addition, glucose sensor failure, sensor noise and difference senseMass metering may be temporarily overlapped, and either be partly overlapped or completely overlapped.Hidden Markov model can be configured at theseEither one or two of part is lower so that microcontroller 32 determines gross mass score61, such as by58 and QacIt is looked between 59Minimum value out.Such as will be discussed below, minimum gross mass score61 can be used in recursive filtering by microcontroller 32In device so as to minimize glucose sensor failure, glucose sensor noise or difference sensing quality influence, even if so as toThe actual glucose that people is provided in the case where there are glucose sensor failure, sensor noise and/or difference sensing quality is horizontalAccurate estimation.
Referring again to FIGS. 4, the gain control module 43 may include recursion filter 52, it can be used to estimation people'sGlucose level.As depicted, until the input of recursion filter 52 is the output from probability analysis tool 54, i.e., by qualityMeasurement group is resulting (i.e. by based on the Glucose results 40 measured58 and be based on corresponding A C impedance measurements 41QAC59 is resulting) minimum gross mass score61.The example for the recursion filter that can be used includes KalmanFilter and extended Kalman filter (EKF).It is of course also possible to use many other types of recursion filter.
In one embodiment, which can be Kalman filter (hereinafter to " Kalman filteringThe reference of device " is also applied for " extended Kalman filter "), it is configured to handle the Portugal measured in second-order linear systemGrape sugar result 40(, that is, raw glucose sensing data), as it is following embodied in the equations.Kalman filter can be outstandingIt includes state vector, indicates the state of the estimation of estimative variable, it is the glucose level of people in this example.CardThalmann filter may include prediction steps (predicting prior state and covariance wherein) and measuring process is (wherein moreNew posteriority kalman gain (Kk), state vector and covariance).Can when receiving new input every time all more new statesVector (i.e. recursively).In the disclosure, state vectorxIn variable can indicate based on the Glucose results 40 measuredPeople actual glucose level estimation.The glucose level vector of estimationxIt can indicate the glucose level of the estimation of peopleg；Its first derivative；And its second dervative.The Glucose results vector measuredzMay include current CG andΔCGValue.?Other dynamic models can be used.VectorxWithzIt can be represented asWith,HerekIndicate thekA sample.Following equation can be used to glucose estimator horizontal vectorx:, hereinkIndicate thekA sample,It is kalman gain, andGross mass score61.In this way, by based on the Glucose results 40 measured58 and be based on corresponding A C impedanceMeasurement result 41QAC59 resulting gross mass scores61 can be used to the Glucose results weighting to measuring,This is in matrixMiddle embodiment.Matrix and support equation for Kalman filter can be such that
, and
。
It would be recognized that parameterIt is to be collected in last sampling periods (such as last 10 minutes) by CGM system 10The raw measurement results (z) of glucose sensor 16 and the filtered measurement result (x) of glucose sensor 16 betweenDifference standard deviation.In addition, parameterIt is the minimum variance for the CGM data collected.Because of glucose sensor 47Always without zero uncertainty, so the parameterThe uncertainty for ensuring to have minimum always.
MatrixAIn parameterWithIt can be configured to be slightly smaller than one (1), so that working as sensor faultWhen inhibit estimation glucose level.MatrixQIt can indicate process noise covariance, andKkIt can indicate that Kalman filter increasesBenefit, use are based on61 and estimateRkTo provide the probability based on controller gain.In addition, the output of filter is (i.e.It is based on61 and estimateRk) be also used as in jointly owned 14/677,148 He of U.S. Patent Application Serial NumberThe input of algorithm described in 14/229,016, disclosures of which are incorporated by reference in here.These algorithmsEach of the probabilistic estimation provided by Kalman filter is all provided, and therefore will not be known by sensorThe influence of the improved estimation of property.The initial estimation for these parameters can be determined as known in the art.
In extended Kalman filter (EKF), nonlinear model can useTo indicateThe system, and also utilize nonlinear modelTo indicate to measure.The nonlinear model may include from otherSourceInput, other sources may include canteen, insulin, take exercise or will affect glucose model other input.This is non-Linear model can be exported from proprietary glucose physiological models.Prediction steps are completed by assessment nonlinear model, and are madeThe Jacobian of the stateful vector field homoemorphism type of apparatusFkTo calculate the uncertainty of prediction.This creation is about current system shapeThe localized linear model of state.It would be recognized thatHkIt is mapping of the kalman filter state (x) between measurement result (z),Wherein filter status (x) includes at least glucose, change rate and acceleration, and measurement result vector (z) only includes glucoseAnd change rate.Following equation can be used by EKF:
。
After the prediction step, glucose present sensor measurement can be used in aligning step.For example,The probability that Kalman filter is configured to controller gain weights the Glucose results currently measured.Such as when61 for it is low when, influence of the Glucose results currently measured to Kalman filter may be close to zero；On the contrary, working as61 for it is high when, the influence of the Glucose results currently measured may be higher.It uses in this way61 can be pairThe logic Modification of the operation of Kalman filter because when sensor fault in the presence of substantial noise occurs orWhen sensor inadequately contacts PWD 11, the Glucose results currently measured may provide the actual glucose water about peopleFlat rarely useful information or no useful information.
Difference sensor fault, sensor noise and sensing quality can promote to estimate the glucose level of people, Yi JizhaoThis Kalman filter can differentially handle them.Sensor noise for normal distribution is attributed to bad sensing and putsIt sets or the reduced perceived quality with the contact of PWD 11, Kalman filter may be configured to find out being averaged for these noisesValue and reduced sensing quality.It may be for each type and/or batch that this, which may be attributed to sensor noise/quality,Glucose sensor and there is the fact that feature, including but not limited in the frequency range of noise and the Glucose results measuredImpedance/amplitude variations correspondence range.These noises and mass property can be some in the parameter of Kalman filterOr it is all it is middle embody (such asOrMiddle embodiment) so that Kalman filter is configured to filter out noise and differenceSensing value, and provide the glucose level of the relatively accurate estimation of people, or even there are the sensings of sensor noise and/or differenceIn the case where quality.On the other hand, sensor fault error is not usually normal distribution, so should be in Kalman filteringIt is differentially processed in device frame.In one embodiment of Kalman filter,61(is true by probability analysis toolIt is fixed) can by Kalman filter using come to the Glucose results weighting measured so that when sensor fault occurs or working asDue to when the contact indicated in counterpart impedance data 41/placement problem sensor mass is bad, the Glucose results that measure40 are largely ignored.
One example of the operation of Kalman filter is shown in Fig. 6, describe the Glucose results 40 measured and is come fromThe output of recursion filter 52, that is, the glucose level 60 of the people estimated.Normally, the glucose level 60 of the estimation of people is usualThe Glucose results 40 measured can be followed.However, during the period 70, sensor may break down；Meanwhile gross massScore61 may be decreased to 0% neighbouring (as determined by the operation as probability analysis tool 54) to consider to control gainIn glucose sensor accuracy low probability.Therefore, Kalman filter can be considered due to low58 it is totalQuality score61 are estimating people in the Portugal of 70 period of period of sensor fault to mitigate the Glucose results measuredImportance in grape sugar level.
With continued reference to Fig. 6, the dextrose equivalent 40 measured may be embodied in period 72 during noise and/or a segment differenceSense quality.Kalman filter can filter the sensing quality of the noise and/or difference to generate during the period 72The glucose level 60 of the estimation of relative smooth.Although the Glucose results measured may be embodied in making an uproar for 72 period of periodThe sensing quality of sound and/or difference, but based on high58 orQac59, gross mass score during this time period61 can be withStill relatively high (such as near 100%) because probability analysis tool can distinguishing sensor noise/sensing quality andSensor fault.As such, what Kalman filter can continue to measure during relatively high importance is placed at the period 72Glucose results are (such as by gross mass score61 is relatively high proved during the period 72).
Glucose sensor measuring uncertainty RkNot usually constant.Currently it can be estimated as nearest sensor surveyMeasure resultz；Gross mass score61；The maximum uncertainty of measurement result；And with continuous glucose sensing knotThe associated normal uncertainty of fruitFunction.ParameterIt can be calculated as controlling the people of not good enough diabetesGlucose maximum physiology variance.It can estimate from the sample of CGM data.Similarly, parameterIt is when appropriate workFor the minimum uncertainty of glucose sensor when making.It can be the optimal cases performance of sensor and when sensor is managedIt can be estimated by the variance of the Glucose results that measure compared with finger prick data when executing with thinking.There may be used forEstimate the other methods of measuring uncertainty, the glucose sensor data including using upper frequency.This can be interpretedThe variance of difference between nearest past CG measurement result and the kalman filter state of estimation.
The glucose level of the estimation of the people such as determined by recursion filter can be used to prediction people in certain following a period of timeBetween glucose level.These estimations can also be used to the behavior and glucose mode of analysis people.Referring back to Fig. 4, predictionAlgorithm 62 can be used to prediction people whether and/or when become hypoglycemia and associated alarm or police can be providedIt accuses.The prediction algorithm 62 from the glucose level 60 of the estimation of 52 recipient of recursion filter and can also receive estimationGlucose level uncertainty.However, it is possible to utilize other input datas (including meal time, carbohydrate, medicineObject, exercise, insulin dose, etc.) enhance the prediction algorithm 62.The prediction algorithm 62 can also be further from other dataSource (Glucose results (i.e. raw glucose sensing data) such as measured or treated glucose sensor data)To receive information.The prediction algorithm 62 can be used Gaussian process and return to learn as indicated by the training pattern 64 in Fig. 4The specific prediction model of patient.The prediction algorithm 62 can also estimate prediction uncertainty, can permit alert threshold becauseIt is adjusted for susceptibility.Alert threshold can also be adjusted based on the current active of people；For example, when people's just sensitivity during sleepDegree may be increased.
For example, it is low to complete that Kalman filter or extension or Kalman filter system model can be usedThe prediction of blood glucose disease.In the example of prediction steps,OrIt is iterated up to desired predictionTime and predicted value is compared with specific threshold value.For example, being predicted if Kalman filter is updated once by per minuteStep can make Kalman filter iteration 45 times to predict from present to the glucose of the 45 minutes following peopleIt is horizontal.Prediction model may include additional prediction input, such as estimated canteen, insulin, exercises or other it is expected that notTo input.
In another example, the estimated dextrose equivalent such as estimated by recursion filterWith the variation speed of dextrose equivalentRateIt is used to define the linear prediction compared with hypoglycemia threshold value.Using following equation by by derivative multiplied by desired pre-Survey the timeTo calculate the dextrose equivalent of predictionTo complete to predict；
。
For example, used specific input vector may include the time t=0, -15 and -30 minutes obtain estimationThree samples of glucose level (CG), estimation glucose level current derivative and t=15 minute derivative, withAnd the time from last canteen.Canteen informationtmealWith pill informationBIt is optional and also may include other data.This can mathematically be expressed as
。
Gaussian process return following equation can be used be based on by (X, y) indicate training data and test point (x*,Y*) predict that the future glucose of people is horizontal:
K (x, x) is covariance function herein.Gauss covariance function can be used to generate as a result, still herein may be usedTo use other functions.The Gauss covariance function that can be used is:
。
The operation of Fig. 7 description prediction algorithm.The Glucose results 40 measured and estimation from glucose sensorGlucose level 60(, that is, Kalman filter output of people) it is illustrated in left side (from time t=- 40 to 0).Current time ist=0.Prediction algorithm can determine the glucose level in the prediction of following sometime (any time for the being greater than t=0) people80.In addition, prediction algorithm can be used to prediction people glucose level whether and/or when can become hypoglycemia.Hypoglycemia threshold value 82 can be established for people, so that actual glucose level means that people has changed into hypoglycemia lower than the threshold value's.Hypoglycemia threshold value 82 can be uniquely determined for everyone.For average people, the threshold value can be about 50 mg/ml.Moreover,It can change for everyone hypoglycemia threshold value 82, so that the threshold value is based on time, event or combinations thereof.For example, rightCan depend on the time on the same day, whether people takes drugs, whether glucose sensor is located in the hypoglycemia threshold value 82 of a peopleIn the state of falling off and/or in dropout state how long etc..The prediction algorithm can predict that people when mayBecome hypoglycemia.In Fig. 7, prediction algorithm can predict people by t=45(i.e. from the 45 of current time minutes) become low bloodSugar.Certainly, as time goes by, prediction algorithm can continue to use the glucose level of nearest estimation (from KalmanFilter) and correspondingly adjust the glucose level of prediction.
Other than it can predict the future value of glucose level of people, which can be further configured to reallyFixed prediction is accurate probability.For example, only prediction in following one or two minute may be pin-point accuracy, and future 60 or 70The prediction of minute may be relatively inaccurate.Certainly, it may be a continuum that prediction, which is accurate probability, forThe 100% of immediate future nearby starts and as prediction will decay near 0% further to future.The information can be used forHypoglycemia warning system is provided in conjunction with actual prediction itself for people.As shown in Figure 7, when 80 foot of the glucose level of predictionWhen being enough higher than hypoglycemia threshold value 82, which may not provide alarm 84,；When the glucose level 80 of prediction is close to low PortugalIt may notify attention 86 when in the preset range of grape sugar threshold value 82；And when the glucose level 80 of prediction drops to glucoseIt may notify danger 88 when 82 or less threshold value.
As discussed previously, which may include the training function for learning the concrete property of people.The training letterNumber can produce the training data that can be used and can be weighted in prediction algorithm based on their influences to generation prediction.The influence level of training data can be by the covariance function k (x, x) that uses in Gaussian process regressor.
Training example or non-training exemplary spanning set be can use to initialize prediction algorithm.When new data are measuredWhen, they can be integrated into prediction algorithm and/or training function.In the presence of for many possible algorithms including new data.These include when 1) predetermined time period be pass by 2) to the prediction of failure of specific data, 3) input data is not indicated onIn training set or 4) it when patient or healthcare provider are included in data (including all new datas) by hand if appropriate, will countAccording to being added to training set.
When being added to training set, new data can be included as new vector, or by existing training vector weightNew weighting.Second method includes keeping constant the benefit of memory needs.After adding additional data, prediction algorithm can beIt is instantly updated in equipment, updates to being reviewed property or updated to being reviewed property at clinic on a personal computer.
With reference to Fig. 4 and 8, the glucose level for estimating to suffer from person with diabetes is shown and is passed for the therapy connectedThe method 100 for sending equipment 31 to select open loop or closed-loop control.Pass through term " connection ", it means that as known in the artWired or wireless connection.This method 100 may include many movements, they can be performed in any suitable order.?In Fig. 8, at movement 102, this method 100 can be by multiple grapes measured from the glucose sensor 16 for being coupled to peopleSugared result 40 and corresponding impedance value 41 receive in Blood sugar management equipment 26.At movement 104, this method 100 be can be usedBlood sugar management equipment 26 is configured to utilize based on the Glucose results 40 measured received in this way and associated resistanceAnti- data 41 determine the probability of sensor mass (i.e. according to the gross mass score of the quality metric of one group of calculating61)Probability analysis tool 54 analyzes multiple Glucose results 40 and corresponding impedance value 41 measured.It, should at movement 106Blood sugar management equipment 26 can be used to use and be configured to utilize gross mass score in method 10061 pairs multiple to measureThe recursion filter of Glucose results weighting estimates the glucose level of people.In addition, at movement 108, this method 100 can be withThe gross mass score obtained using Blood sugar management equipment 26 based on result61 selections operate therapy delivery device wherein31 open loop glucose control 45 or closed loop glucose control 47.Can as described above as establish probability analysis workTool and recursion filter.
It should now be understood that method and system described herein can be used to the grape that person with diabetes is suffered from estimationSugar level, even if there are noise and/or sensor inaccuracies (such as sensor signal loss), and in given phase periodBetween based on conventional receiver to the Glucose results 40 that measure and associated impedance data 41 automatically select microcontroller32 will operate the operation mode (open loop or closed loop) of therapy delivery device 31.In addition, method and system described herein may be used alsoIt is horizontal with the future glucose for being used to prediction people.As such, they can predict whether the glucose level of people may becomeWhen hypoglycemia is likely to become at hypoglycemia and/or the glucose level of people.Hypoglycemia is likely to become detecting or predicting peopleWhen, this method and system can provide corresponding information to people, such as.Method described herein can be depositedStoring up has the computer-readable medium for executing the computer executable instructions of method.Such computer-readable medium canTo include compact disk, hard disk drive, thumb actuator, random access memory, dynamic random access memory, flash storageDevice, etc..
It is to be noted, that such as opposite with the detailed description of desired use, " being configured " in a specific way, " being configured " is special at embodyingThe detailed description herein of the component of qualitative matter or the present disclosure to work in a specific way is the detailed description in structure.MoreSpecifically, the existing physical state of component is indicated the reference of the mode of component " being configured " herein, and to be adopted like thisIt is used as the clearly detailed description of the architectural characteristic of component.
Although described herein and describe the particular embodiment of the present invention and aspect, can without departing fromVarious other changes and modification are made in the case where the spirit and scope of the present invention.In addition, although being described hereinVarious creative aspects, but do not need to combine such aspect and utilize.Therefore it is intended to be covered on appended claimsAll such changes and modification in the scope of the present invention.